Identity Is Not Authority: The Missing Boundary in Agentic AI
Year: 2026
Type: article
Abstract: Identity, authentication, and authorization do not by themselves establish legitimate execution authority in agentic AI. AI agents and autonomous systems require a separate authority boundary that determines what a system may do, when execution must be constrained or refused, and how authority changes under degraded or uncertain conditions. Agentic AI, AI agents, and autonomous agents increasingly combine identity, authentication, authorization, access control, autonomous planning, tool use, policy enforcement, and runtime execution. As these systems gain greater machine autonomy, a critical governance question emerges: does identifying and authenticating an AI agent establish that the agent has authority to act? This paper examines the missing architectural distinction between identity, access control, capability, and execution authority in autonomous systems, distributed systems, and AI-driven infrastructure, autonomous systems, AI governance, runtime governance, AI authorization, access control, and execution control. Identity, authentication, credentials, roles, permissions, and technical capability may establish who or what an agent is and what resources it can reach, but they do not independently establish the decision authority or execution authority required for a particular action under current conditions. The paper introduces the Stable Authority Boundary (SAB) as a pre-execution trust boundary at which proposed execution is evaluated against recognized authority. Rather than treating prior authentication, access, permission, or capability as sufficient authorization for consequential action, SAB requires the applicable authority state, scope, conditions, and evidence to be evaluated before execution proceeds. This distinction connects AI authorization and runtime authorization to runtime governance, policy enforcement, authorized execution, runtime assurance, and execution governance. An autonomous agent may remain technically capable of acting while the conditions supporting legitimate execution have changed. Under those circumstances, execution may need to be constrained, held, escalated, or refused. The architecture is relevant to agentic AI, AI agents, autonomous agents, autonomous systems, machine autonomy, AI governance, runtime governance, AI authorization, authentication and authorization, access control, policy enforcement, trust boundaries, runtime assurance, tool authorization, tool use, decision authority, execution authority, and authorized execution. The central distinction is simple: identity establishes who or what is acting; authority determines whether the proposed execution may legitimately proceed now. BLOCK VECTOR Research ArchitectureThis Research Map is the navigation surface for the BLOCK VECTOR publication architecture. The interactive version is maintained at: https://blockvectortech.com/research-map.html Archival Research Map — Version 1.0: https://doi.org/10.5281/zenodo.22102620 Individual publications within the architecture link back to this map, while this map provides direct routes to the current publication set.
Author: Forbes, David
Institution: Vector (United States)
Language: English
Topic: Access Control and Trust
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7204475552